|
Informatics: Logistic Regression: A Self-Learning Text
|
|---|
David G. Kleinbaum. 282 pages. New York: Springer-Verlag; 1994. $49.00.
Logistic regression models are widely used in medical research today. Kleinbaum illustrates how logistic regression measures the strength of the association (odds ratio) between exposure and disease outcome adjusted for confounding variables in models with and without statistical interactions. By presenting the basic concepts of odds ratio, confounding, and interaction in many detailed examples before examining the more complicated issues of statistical inference, the author has created a comprehensible introduction. Although the probability of a person developing a disease (risk) can be obtained from statistical software packages for studies of any design, Kleinbaum explains that it is only appropriate to use risk when interpreting cohort (longitudinal) studies, not when interpreting case–control or cross-sectional studies.
Many features make this text more useful for beginners than previous texts. Answers to straightforward questions provide rapid feedback. Detailed content outlines with page indications help the reader locate material on areas found to be difficult, making the book a useful reference.
Although odds ratios are more clearly distinguished from risk ratios here than in most presentations, in an answer to an exercise, an odds ratio of 1.48 was interpreted incorrectly as "1.48 times as likely" (page 268) rather than as 1.48 times the odds. Further, using the term "risk odds ratio" as a type of odds ratio (not a risk ratio) is confusing. Perhaps those problems and a few typographical errors will be addressed in subsequent editions.
Although many nonelementary topics, such as multicolinearity and regression diagnostics, are beyond its scope, this text does provide some materials useful for teachers and more advanced readers. These include computer data sets, pitfalls to avoid when modeling interactions, information on when to use conditional and unconditional logistic regression, and modeling strategy guidelines that also apply to multiple linear regression and Cox regression. Because of its emphasis on commonalities in statistical modeling and its clarity, Logistic Regression is an excellent place to begin (or continue) the study of multivariable statistical modeling in medicine.
Article
